[mellanox|ffb] use system level warm reboot for Mellanox fastfast boot (#2374)

* [mellanox|ffb] use system level warm reboot for Mellanox fastfast boot

Signed-off-by: Stepan Blyschak <stepanb@mellanox.com>

* [mellanox|ffb] add comments for mellanox start/stop drivers section

Signed-off-by: Stepan Blyschak <stepanb@mellanox.com>

files/build_templates/docker_image_ctl.j2

@@ -28,7 +28,7 @@ function preStartAction()
 {
 {%- if docker_container_name == "database" %}
     WARM_DIR=/host/warmboot
-    if [[ "$BOOT_TYPE" == "warm" && -f $WARM_DIR/dump.rdb ]]; then
+    if [[ ("$BOOT_TYPE" == "warm" || "$BOOT_TYPE" == "fastfast")  && -f $WARM_DIR/dump.rdb ]]; then
         # Load redis content from /host/warmboot/dump.rdb
         docker cp $WARM_DIR/dump.rdb database:/var/lib/redis/dump.rdb
     else
@@ -49,7 +49,7 @@ function postStartAction()
     until [[ $(/usr/bin/docker exec database redis-cli -s $REDIS_SOCK ping | grep -c PONG) -gt 0 ]]; do
         sleep 1;
     done
-    if [[ "$BOOT_TYPE" == "warm" && -f $WARM_DIR/dump.rdb ]]; then
+    if [[ ("$BOOT_TYPE" == "warm" || "$BOOT_TYPE" == "fastfast") && -f $WARM_DIR/dump.rdb ]]; then
         rm -f $WARM_DIR/dump.rdb
     else
         # If there is a config db dump file, load it
@@ -61,7 +61,7 @@ function postStartAction()
     fi
 {%- elif docker_container_name == "swss" %}
     docker exec swss rm -f /ready   # remove cruft
-    if [[ "$BOOT_TYPE" == "fast" || "$BOOT_TYPE" == "fastfast" ]] && [[ -d /host/fast-reboot ]]; then
+    if [[ "$BOOT_TYPE" == "fast" ]] && [[ -d /host/fast-reboot ]]; then
         test -e /host/fast-reboot/fdb.json && docker cp /host/fast-reboot/fdb.json swss:/
         test -e /host/fast-reboot/arp.json && docker cp /host/fast-reboot/arp.json swss:/
         test -e /host/fast-reboot/default_routes.json && docker cp /host/fast-reboot/default_routes.json swss:/

files/scripts/swss.sh

@@ -90,11 +90,7 @@ start() {
 
     # Don't flush DB during warm boot
     if [[ x"$WARM_BOOT" != x"true" ]]; then
-        # Don't flush APP_DB during MLNX fastfast boot
+        /usr/bin/docker exec database redis-cli -n 0 FLUSHDB
-        BOOT_TYPE="$(cat /proc/cmdline | grep -o 'SONIC_BOOT_TYPE=\S*' | cut -d'=' -f2)"
-        if [[ x"$BOOT_TYPE" != x"fastfast" ]] && [[ ! -f /var/warmboot/issu_started ]]; then
-            /usr/bin/docker exec database redis-cli -n 0 FLUSHDB
-        fi
         /usr/bin/docker exec database redis-cli -n 2 FLUSHDB
         /usr/bin/docker exec database redis-cli -n 5 FLUSHDB
         clean_up_tables 6 "'PORT_TABLE*', 'MGMT_PORT_TABLE*', 'VLAN_TABLE*', 'VLAN_MEMBER_TABLE*', 'INTERFACE_TABLE*', 'MIRROR_SESSION*'"

files/scripts/syncd.sh

@@ -90,14 +90,12 @@ start() {
 
         # Flush DB during non-warm start
         /usr/bin/docker exec database redis-cli -n 1 FLUSHDB
-
-        # platform specific tasks
-        if [ x$sonic_asic_platform == x'cavium' ]; then
-            /etc/init.d/xpnet.sh start
-        fi
     fi
 
     # platform specific tasks
+
+    # start mellanox drivers regardless of
+    # boot type
     if [ x"$sonic_asic_platform" == x"mellanox" ]; then
         BOOT_TYPE=`getBootType`
         if [[ x"$WARM_BOOT" == x"true" || x"$BOOT_TYPE" == x"fast" ]]; then
@@ -109,6 +107,13 @@ start() {
         /sbin/modprobe i2c-dev
     fi
 
+    if [[ x"$WARM_BOOT" != x"true" ]]; then
+        if [ x$sonic_asic_platform == x'cavium' ]; then
+            /etc/init.d/xpnet.sh start
+        fi
+    fi
+
+
     # start service docker
     /usr/bin/${SERVICE}.sh start
     debug "Started ${SERVICE} service..."
@@ -146,21 +151,23 @@ stop() {
     /usr/bin/${SERVICE}.sh stop
     debug "Stopped ${SERVICE} service..."
 
-    # if warm start enabled, don't stop peer service docker
+    # platform specific tasks
+
+    # stop mellanox driver regardless of
+    # shutdown type
+    if [ x$sonic_asic_platform == x'mellanox' ]; then
+        /etc/init.d/sxdkernel stop
+        /usr/bin/mst stop
+    fi
+
+
     if [[ x"$WARM_BOOT" != x"true" ]]; then
-        # platform specific tasks
         if [ x$sonic_asic_platform == x'cavium' ]; then
             /etc/init.d/xpnet.sh stop
             /etc/init.d/xpnet.sh start
         fi
     fi
 
-    # platform specific tasks
-    if [ x"$sonic_asic_platform" == x"mellanox" ]; then
-        /etc/init.d/sxdkernel stop
-        /usr/bin/mst stop
-    fi
-
     unlock_service_state_change
 }
 

platform/mellanox/docker-syncd-mlnx/start.sh

@@ -8,8 +8,3 @@ supervisorctl start syncd
 
 supervisorctl start mlnx-sfpd
 
-BOOT_TYPE="$(cat /proc/cmdline | grep -o 'SONIC_BOOT_TYPE=\S*' | cut -d'=' -f2)"
-if [[ x"$BOOT_TYPE" == x"fastfast" ]] && [[ -f /var/warmboot/issu_started ]]; then
-    rm -f /var/warmboot/issu_started
-    /usr/bin/ffb &>/dev/null &
-fi

platform/mellanox/mlnx-ffb.sh

@@ -62,6 +62,19 @@ check_sdk_upgrade()
     return "${CHECK_RESULT}"
 }
 
+check_ffb()
+{
+    check_issu_enabled || {
+        echo "ISSU is not enabled on this HWSKU"
+        return "${FFB_FAILURE}"
+    }
+    check_sdk_upgrade || {
+        echo "SDK upgrade check failued"
+        return "${FFB_FAILURE}"
+    }
+    return "${FFB_SUCCESS}";
+}
+
 # Perform ISSU start
 issu_start()
 {
@@ -70,8 +83,6 @@ issu_start()
 
     EXIT_CODE=$?
 
-    touch /host/warmboot/issu_started
-
     return $EXIT_CODE
 }
 

platform/mellanox/mlnx-issu/scripts/ffb

@@ -1,69 +0,0 @@
-#!/usr/bin/env python
-"""
-Part of Mellanox platform specific fastfast boot implementation for warm-boot.
-Notifies SYNCD proccess once boot is finished after warm-reboot.
-Once SYNCD received such notification it should set appropriate SAI attribute.
-Then SAI will notify SDK to end ISSU mode for the FFB.
-"""
-
-
-import time
-import swsssdk
-from threading import Timer
-
-
-class FFB(object):
-    """Provides implementation for Mellanox fastfast boot"""
-    DB_WARM_TABLE_KEY = 'WARM_RESTART_TABLE|bgp'
-    DB_STATE_ENTRY_NAME = 'state'
-    DB_STATE_TYPE_RECONCILED = 'reconciled'
-    DB_CHANNEL_NAME = 'MLNX_FFB'
-    DB_CHANNEL_MSG = '["SET","ISSU_END"]' # message should be in the following format: ["<operation>","<data>"]
-    SUB_THREAD_TIMEOUT = 1
-    STOP_TIMER_TIMEOUT = 180
-
-    def __init__(self):
-        self.state_db = swsssdk.SonicV2Connector()
-        self.state_db.connect(self.state_db.STATE_DB)
-
-        self.prevState = self.state_db.get(self.state_db.STATE_DB, self.DB_WARM_TABLE_KEY, self.DB_STATE_ENTRY_NAME)
-
-        self.pubSub = self.state_db.redis_clients[self.state_db.STATE_DB].pubsub()
-        self.pubSub.psubscribe(**{'__key*@6__:{}'.format(self.DB_WARM_TABLE_KEY): self.eventHandler})
-
-        self.timeoutTimer = Timer(self.STOP_TIMER_TIMEOUT, self.finish)
-
-    def run(self):
-        # Start event thread in order to get required events
-        self.eventThread = self.pubSub.run_in_thread(sleep_time=self.SUB_THREAD_TIMEOUT)
-        # Start oneshot timer in order to exit in case required event is not received during defined timeout
-        self.timeoutTimer.start()
-
-    def finish(self):
-        # Stop event thread and timeout timer
-        self.eventThread.stop()
-        self.timeoutTimer.cancel()
-
-        # Publish "FFB END" event to SYNCD process
-        time.sleep(60) # W/A: Wait until configuration is applied to HW since it takes some time
-        self.state_db.publish(self.state_db.STATE_DB, self.DB_CHANNEL_NAME, self.DB_CHANNEL_MSG)
-
-    def eventHandler(self, msg):
-        # Only "set" operations are needed so just skip all others
-        if msg['data'] != 'hset':
-            return
-
-        state = self.state_db.get(self.state_db.STATE_DB, self.DB_WARM_TABLE_KEY, self.DB_STATE_ENTRY_NAME)
-
-        if (state != self.prevState) and (state == self.DB_STATE_TYPE_RECONCILED):
-            self.finish()
-        else:
-            self.prevState = state
-
-
-def main():
-    FFB().run()
-
-
-if __name__ == '__main__':
-    main()

platform/mellanox/mlnx-issu/setup.py

@@ -10,6 +10,5 @@ setup(
     maintainer_email='stepanb@mellanox.com',
     scripts=[
         'scripts/issu',
-        'scripts/ffb',
     ]
 )